Texture Image Segmentation Based on the Elements of Gray Level Aura Matrices Zohra Haliche, Kamal Hammouche
Jack-Gerard Postaire
Departement d'Automatique Universite Mouloud Mammeri Tizi-Ouzou, Algerie zohrahaliche@ yahoo.fr kamal hammouche@ yahoo.fr
Laboratoire d' Automatique, Genie Informatique et Signal Universite des Sciences et Technologies de Lille (USTL) Villeneuve d' Ascq, France
[email protected]
Abstract- We present a method for texture image segmentation based on Gray Level Aura Matrices (GLAMs). The GLAMs allow describing
the
relationship
between
the
target
pixel
and
its
neighboring pixels located in neighborhood structure defined by a structuring element. Their elements are directly used in this paper instead of Haralick features in order to characterize each pixel of the image. The pixels having the same features are then gathered into classes using the Fuzzy C-means algorithm.
Experiments
results on synthetic and real images show the relevance of the elements of GLAMs in the
segmentation of images with different
textures.
Keywords-segmentation; texture; Gray Level Aura Matrices.
I.
INTRODUCTION
The segmentation of textured images is one of fundamental problem in computer vision which concerns a wide variety of practical applications such as robotics, remote sensing, medical imaging and so on [1]. Its main goal is to partition an image into homogeneous regions with respect to texture properties where each region corresponding to an object or a part of the objects constituting the real scene. The success of the segmentation depends mainly on the texture features used to characterize the pixels of the image. These features can be defined by different techniques [2]. Among them, the technique based on the Gray Level Co-occurrence Matrices (GLCM) is very popular [3]. Generally, these features yield good results [4-5], although the GLCM reflects only the relationships between neighboring pixels taken two by two. A generalization of the GLCM, called Gray Level Aura Matrix (GLAM), has been proposed in [6]. The GLAM is defmed in a framework based on the set-theoretic concept of "aura set". This matrix allows to quantity the importance of a set of pixels with a specified gray level standing in the neighborhood of another set of pixels having another gray level. The amount of neighboring pixels with the specified gray level is quantified by means of a measure defined on this set: the "aura measure". The GLAM counts, by means of the aura measures, the occurrences of pairs of gray levels associated to a pixel and its neighbors belonging to a neighborhood defined by a structuring element.
The GLAMs and its variants are used as tools for the texture representation [7], the image retrieval [8, 9], the texture synthesis [10, 11], the texture classification [12-18] and more recently for the segmentation of textured images [19]. In [19], Haralick features, extracted from the GLAM, are used as texture features. In this article we use directly the elements of the GLAM as texture features instead of the Haralick features. This approach consists to construct a GLAM for each pixel from all pixels located in a neighborhood window. The elements of the GLAM are considered as texture features for each corresponding pixel. The pixels having similar features are then gathered into classes of textures using one of the unsupervised classification methods such that the Fuzzy C-Means algorithm. II.
AURA CONCEPT
A. Aura set We consider an image as a finite lattice S. Let N be a structuring element. When centered on any pixel Ps E S , this structuring element defines a set Ns of neighbors. The gray level of a pixel Ps is denoted I(Ps)' Let S9 and S9' be two subsets of the window S (S9,S9'
